Serum prolactin as a shared inflammation-associated signal across inflammatory breast conditions: a four-group comparative study with exploratory inflammatory breast cancer subtype analysis

Abstract

Background

Inflammatory breast cancer (IBC) and nonpuerperal mastitis (NPM) present with closely overlapping inflammatory breast signs yet require entirely different management. No previous study has reported on the serum prolactin (PRL) level in the context of IBC or compared the PRL level between IBC and NPM. This study aimed to characterize serum PRL in IBC relative to NPM, invasive ductal carcinoma, and healthy controls, and to determine whether PRL adds discriminative information beyond systemic inflammatory markers for distinguishing IBC from NPM.

Methods

A single-center, retrospective, four-group comparison [healthy controls (HC): n=121; invasive ductal carcinoma (IDC): n=123; IBC: n=42; NPM: n=226; total: N=512] was conducted on consecutive patients evaluated between January 2022 and December 2025. Three prespecified contrasts (NPM vs. HC, IBC vs. IDC, and IBC vs. NPM) were analyzed via multivariable logistic regression adjusted for age, body mass index (BMI), and menopausal status. Within-group PRL-inflammation correlations were compared across groups via Fisher r-to-z transformation. A combined model consisting of log(PRL), log[C-reactive protein (CRP)], log(neutrophil-to-lymphocyte ratio) and age, was evaluated for IBC-NPM discrimination.

Results

PRL was elevated in both inflammatory conditions [NPM vs. HC: adjusted odds ratio (OR) 3.88, 95% confidence interval (CI): 2.52–6.20; P<0.001; IBC vs. IDC: adjusted OR 4.03, 95% CI: 2.19–7.97, median PRL 342 vs. 208 mIU/L, P<0.001]—the first report of PRL elevation in patients with IBC. However, PRL could not discriminate IBC from NPM (adjusted OR 0.94; 95% CI: 0.31–2.97; P=0.91). A four-variable model discriminated IBC from NPM (apparent area under the curve [AUC] 0.895; optimism-corrected 0.879), but this was driven by CRP (OR 0.68, P=0.04) and age (OR 1.19, P<0.001), not PRL (P=0.62). In NPM, PRL correlated with interleukin (IL)-6, IL-1β, and tumor necrosis factor (TNF)-α (P<0.001; false-discovery rate [FDR] q<0.001); this coupling was not detectable in the IBC cytokine subset (n=31; Fisher r-to-z for IL-6: P=0.10). Within IBC, hyperprolactinemia (PRL >496 mIU/L; 13/42, 31%) appeared more frequent in luminal B human epidermal growth factor receptor 2 (HER2)+ and triple-negative breast cancer (TNBC) (Monte Carlo P=0.03; post hoc Fisher OR 10.4, P=0.006), with no cases in luminal A (0/8) or HER2-enriched (0/4). However, with only 42 IBC patients and 13 hyperprolactinemic cases, statistical power is severely limited, and this finding is hypothesis-generating only.

Conclusions

Serum PRL is elevated across inflammatory breast conditions and appears to behave as a shared, inflammation-associated endocrine marker rather than a marker that distinguishes IBC from NPM. Serum PRL is elevated across inflammatory breast conditions and behaves as a shared inflammation-associated marker rather than a discriminator between IBC and NPM. Separation between the two conditions was driven by CRP and age. Within IBC, hyperprolactinemia appeared more frequent in aggressive subtypes, but this is strictly an exploratory, hypothesis-generating observation requiring prospective validation in an independent, adequately powered cohort. Because tumor prolactin receptor (PRLR) expression was not measured, any inference regarding PRL/PRLR pathway activation in IBC remains indirect.

Highlight box.

Key findings

• Serum prolactin (PRL) was elevated in inflammatory breast cancer (IBC) and nonpuerperal mastitis (NPM) as compared to controls but could not distinguish IBC from NPM [odds ratio (OR) 0.94; P=0.913], discrimination was driven by C-reactive protein (CRP) and age.

• Hyperprolactinemia (PRL >496 mIU/L) occurred in 13/42 patients with IBC (31%) and appeared more frequent in the luminal B HER2+ and triple-negative breast cancer (TNBC) (specificity 90%; post hoc OR 10.4). It was also correlated with tumor burden (ρ=0.46), but this association was driven by the high-PRL subgroup.

• PRL correlated with serum levels of cytokines in the NPM group but not in the IBC group.

What is known and what is new?

• PRL is implicated in benign mastitis and has been associated with breast cancer risk, but its role in IBC and IBC-NPM discrimination was unexamined.

• This study first characterized PRL in IBC; PRL appears to be a shared inflammation-associated signal rather than a diagnostic discriminator. A hyperprolactinemic subset appeared more frequent in aggressive subtypes, but tumor prolactin receptor (PRLR) expression was not measured, so any inference regarding PRL/PRLR pathway activation is indirect.

What is the implication, and what should change now?

• PRL should not be used to distinguish IBC from NPM, as CRP and age provide the main discriminative information.

• These findings are exploratory and remain to be validated in prospective, large-sample cohorts.

• Hyperprolactinemia may identify a candidate subset for future PRLR-targeted studies; confirmation of tumor PRLR expression and local PRL production is warranted.

Introduction

Inflammatory breast cancer (IBC) is a rare but exceptionally aggressive form of locally advanced breast cancer, accounting for a small proportion of breast cancer diagnoses yet a disproportionate share of breast cancer deaths, with a virulent clinical course and poor survival even with multimodal treatment (1). It is defined clinically rather than by a single histological feature: according to the 8th edition of the American Joint Committee on Cancer (AJCC) Cancer Staging Manual, IBC corresponds to stage T4d; it is characterized by the rapid onset of diffuse erythema and edema (peau d’orange) involving at least one-third of the breast skin, with a duration of less than 6 months from the onset of symptoms to histological confirmation of invasive carcinoma (2,3).

This clinical, inflammation-based presentation is the source of a persistent diagnostic difficulty: IBC closely mimics benign inflammatory breast disease, particularly nonpuerperal mastitis (NPM), an umbrella term for nonspecific inflammatory conditions of the breast in nonpregnant, nonlactating women, including idiopathic granulomatous (lobular) mastitis, plasma cell mastitis, and periductal mastitis (4,5). Both conditions present with breast erythema, edema, pain, and a palpable mass, and early IBC with milder signs may be misdiagnosed as mastitis, delaying appropriate oncological treatment (6). Reliable, rapidly available markers that can inform the early triage of an inflammatory breast presentation would thus have substantial clinical value.

Prolactin (PRL), a pituitary polypeptide hormone, has attracted research attention in both of these contexts, albeit from different perspectives. Hyperprolactinemia has been frequently associated with NPM and idiopathic granulomatous mastitis, and PRL has been suggested to contribute to the pathogenesis and severity of benign inflammatory breast disease (5-9). Beyond its lactogenic role, PRL is now recognized as an immunomodulatory mediator within a bidirectional neuroendocrine-immune network: proinflammatory cytokines [interleukin (IL)-1β, IL-6, and tumor necrosis factor (TNF)-α] can stimulate pituitary PRL release, while PRL in turn acts on immune cells through the PRL receptor (PRLR), and hyperprolactinemia recurs in the active phase of a variety of inflammatory and autoimmune diseases (10-12). Notably, PRL is also produced at extrapituitary sites, including the breast epithelium itself as well as activated leukocytes and fibroblasts, where it can act in autocrine or paracrine fashion (11,13,14). Additionally, PRL signaling through the PRLR has long been implicated in breast tumorigenesis, acting through the JAK2-STAT5, PI3K-AKT, and RAS-MAPK cascades that drive proliferation and survival in mammary epithelial and breast cancer cells (14-18). Consistent with this, higher circulating PRL has been associated with an increased risk of breast cancer in large prospective cohorts and meta-analyses, an association strongest for patients with postmenopausal and hormone receptor-positive disease (19,20). Moreover, elevated serum PRL has been linked to poorer response to endocrine therapy and shorter survival in patients with advanced breast cancer (21). Recent evidence further suggests that low baseline serum PRL levels are associated with improved progression-free survival in individuals with breast cancer (22). Despite evidence linking PRL both to benign inflammatory breast disease and to breast cancer risk and progression, serum PRL has, to our knowledge, not been examined regarding its role in IBC, nor has PRL level been directly compared between IBC and NPM—the two conditions in which an inflammation-associated endocrine signal may be highly clinically relevant.

We therefore designed a four-group comparative study, including healthy control (HC), invasive ductal carcinoma (IDC), IBC, and NPM groups, to locate IBC within a structured framework of healthy, benign-inflammatory, noninflammatory-malignant, and inflammatory-malignant states. We hypothesized that serum PRL is elevated across inflammatory breast conditions (both NPM and IBC) relative to their noninflammatory comparators and represents a shared inflammation-associated endocrine state rather than a disease-specific diagnostic signal. Specifically, we predicted that PRL would distinguish each inflammatory condition from its own control (i.e., NPM from HC and IBC from IDC) but could not distinguish the two inflammatory conditions from each other and that any separation achievable between IBC and NPM would be derived from the magnitude of systemic inflammation [chiefly C-reactive protein (CRP)] rather than from PRL itself. We frame these relationships within an inflammation-pituitary-prolactin (IPP) axis (10,12,23,24). We present this article in accordance with the TRIPOD reporting checklist (available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0582/rc).

Methods

Study design, setting, and participants

A single-center, retrospective, cross-sectional study was conducted to evaluate the diagnostic and biological role of serum PRL across four breast conditions with overlapping inflammatory or clinical presentations represented in the HC, IDC, IBC, and NPM groups. Between January 2022 and December 2025, consecutive women evaluated at the Breast Surgery Department of Yueyang Hospital of Integrated Traditional Chinese and Western Medicine were retrospectively screened for eligibility. Clinical, laboratory, imaging, and pathological data were retrieved from the institutional electronic medical record and laboratory information system. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Yueyang Hospital of Integrated Traditional Chinese and Western Medicine, Shanghai University of Traditional Chinese Medicine (approval No. 2026-104). The requirement for written informed consent was waived due to the retrospective nature of the analysis and the use of anonymized routine clinical data.

After the application of the eligibility criteria (described below), a total of 512 women were included in the final analysis, comprising 121 HCs, 123 patients with IDC, 42 patients with IBC, and 226 patients with NPM. All participants were nonpregnant and nonlactating women. The four groups were created to permit a structured, layered set of comparisons rather than a single case-control contrast. HC and IDC served as the noninflammatory reference conditions (a physiologically normal state and a noninflammatory malignancy, respectively), whereas NPM and IBC constituted the two clinically critical “inflammatory” phenotypes that frequently mimic one another at presentation (6).

Diagnostic criteria and group definitions

IBC

IBC was diagnosed according to the AJCC 8th edition criteria for stage T4d disease, defined as a clinical-pathological entity characterized by the rapid onset of diffuse erythema and edema (peau d’orange) involving at least one-third of the breast skin, with a time from first symptom to diagnosis of no more than 6 months, together with histopathological confirmation of invasive carcinoma on core-needle or skin biopsy (2,3). The presence of dermal lymphovascular tumor emboli supported but were not required for the diagnosis, consistent with the AJCC definition and with proposed consensus diagnostic criteria for IBC (6).

IDC

The IDC group comprised women with histologically confirmed invasive carcinoma of no special type without the clinical features of IBC, serving as a noninflammatory malignant comparator.

NPM

NPM was diagnosed on the basis of clinical presentation (breast mass, pain, erythema, nipple retraction or discharge, presence of abscess, sinus, or fistula formation) together with histopathological confirmation of a nonspecific inflammatory pattern consistent with idiopathic granulomatous (lobular) mastitis, plasma cell mastitis, or periductal mastitis, in accordance with established diagnostic criteria for NPM in nonpregnant, nonlactating women (4,5). Cases with a confirmed specific infectious etiology (e.g., tuberculous mastitis) were excluded.

HCs

HCs were women without breast malignancy or active breast inflammatory disease who underwent routine breast health evaluation, with peripheral blood and endocrine measurements available.

Exclusion criteria

Across all groups, women were excluded if they were pregnant or lactating at presentation, or if they had a documented condition or were receiving a medication known to independently alter serum PRL or systemic inflammatory indices (because such factors could confound the interpretation of PRL as an inflammation-associated signal) (25). Prespecified PRL-confounding factors, screened for in every candidate participant, including the following: (I) a known pituitary adenoma or prolactinoma or other sellar/hypothalamic lesion; (II) primary hypothyroidism (which raises PRL through elevated thyrotropin-releasing hormone); (III) chronic renal failure or significant hepatic insufficiency (which reduce PRL clearance); (IV) macroprolactinemia (where documented); and (V) concurrent use of PRL-elevating drugs, including dopamine-receptor antagonists (typical and atypical antipsychotics, metoclopramide, and domperidone), selected antidepressants, verapamil, estrogen-containing preparations, and opioids. These factors were ascertained by structured review of the electronic medical record (including problem list, encounter diagnoses, and the active medication list) supplemented, for the malignant groups, by the pretreatment anesthetic and oncology assessments; when the record was ambiguous, the case was adjudicated by two reviewers. Women were additionally excluded if they had a concurrent active infection unrelated to the breast, another malignancy, or another acute systemic inflammatory illness that could independently elevate the inflammatory indices. Finally, participants with incomplete records for the core variables required for the primary analyses (serum PRL and the principal peripheral blood markers) were not included in the final dataset. The number of women screened, the number excluded at each step with reasons, and the number analyzed in each group are reported in the participant-flow diagram in Figure 1 (group-specific exclusion counts are detailed in this figure). Outcome status (IBC, IDC, or NPM) was determined by histopathology and clinical records, independent of and with blinding to serum PRL measurements.

Figure 1.

Flowchart of participant inclusion and study design (STROBE diagram). The flowchart shows the female participants through screening, eligibility, and analysis across the four-group comparative design (HCs, IDC, IBC, and NPM; N=512). The diagram shows, at each stage, the number screened, the number excluded with reasons (prespecified prolactin-confounding conditions and medications, pregnancy or lactation, concurrent infection or other systemic inflammatory illness, and missing core data), and the number analyzed in each group; bracketed values are placeholders for the screened, excluded, and per-group exclusion counts to be inserted from the study records. The four groups position IBC within a framework spanning healthy, benign-inflammatory, noninflammatory-malignant, and inflammatory-malignant states, supporting the prespecified three-layer comparison structure (layer 1: NPM vs. HC; layer 2: IBC vs. IDC; layer 3: IBC vs. NPM). *, pituitary adenoma/prolactinoma, primary hypothyroidism, renal or hepatic insufficiency, macroprolactinemia, or prolactin-elevating drugs; †, serum PRL and the principal peripheral blood markers required for the primary analyses. HC, healthy control; IBC, inflammatory breast cancer; IDC, invasive ductal carcinoma; NPM, nonpuerperal mastitis; PRL, prolactin; STROBE, Strengthening the Reporting of Observational Studies in Epidemiology.

Data collection and variables

For each participant, we extracted demographic and clinical variables [age, body mass index (BMI), and menopausal status] and a panel of peripheral blood and endocrine markers measured at presentation, before the initiation of disease-specific treatment. The endocrine marker of primary interest was serum PRL (mIU/L). The inflammatory and hematological panel comprised CRP, white blood cell count (WBC), neutrophil count, lymphocyte count, monocyte count, and platelet count, from which composite indices were derived: the neutrophil-to-lymphocyte ratio (NLR), platelet-to-lymphocyte ratio (PLR), monocyte-to-lymphocyte ratio (MLR), and the systemic immune-inflammation index (SII; calculated as platelet count × neutrophil count/lymphocyte count) (26). For the IDC group, BMI was computed from recorded height and weight where the BMI field was not directly populated in the source dataset.

Data from serum proinflammatory cytokines (IL-6, IL-1β, and TNF-α) were retrieved when available; these had been measured for all HC and NPM participants and for subsets of the IBC (n=31) and IDC (n=56) groups. All laboratory values were used as recorded in the source laboratory information system, with the assays and reference ranges of the central clinical laboratory being used. In the IBC group, pretreatment tumor burden was characterized according to the maximum tumor diameter measured on breast ultrasound (n=42; all patients in the group) and magnetic resonance imaging (MRI; n=28). Because most patients with IBC received neoadjuvant systemic therapy that markedly reduced lesion size before surgery, pretreatment imaging measurements rather than posttreatment pathological dimensions were used as the primary tumor-burden indicator. Additional pathological variables recorded for the IBC group included estrogen receptor (ER) and progesterone receptor (PR) expression (%), human epidermal growth factor receptor 2 (HER2) status, Ki-67 index, and the presence of nerve involvement and lymphovascular invasion. Laboratory technicians performing the assays were blinded to the participants' clinical groups and outcomes.

Molecular subtype classification

Within the IBC group, tumors were assigned to molecular subtypes according to immunohistochemistry (IHC)-based surrogate definitions delineated by the St. Gallen consensus criteria and included luminal A, luminal B HER2−, luminal B HER2+, HER2-enriched, and triple-negative breast cancer (TNBC) (27). Receptor status was determined by routine IHC, and HER2 positivity was defined as an IHC score of 3+ or 2+ with fluorescence in situ hybridization amplification. Hyperprolactinemia was defined a priori according to the clinical laboratory upper limit of normal for nonpregnant women (serum PRL >496 mIU/L).

Statistical analysis

Descriptive statistics and group comparisons

Continuous variables are expressed as the median and interquartile range (IQR) and were compared across the four groups via the Kruskal-Wallis test; categorical variables are expressed as counts (percentages) and were compared via the chi-square test. Skewed biomarkers were log-transformed for parametric modelling and for visualization on logarithmic axes.

Phenotype mapping and three-layer comparison

To visualize the multidimensional inflammatory-endocrine landscape, serum PRL was plotted against CRP in two-dimensional log-log space with group centroids and 95% confidence ellipses, and an exploratory principal component analysis (PCA) was performed on seven log-transformed markers and age after standardization. The diagnostic information carried by PRL was dissected through the following prespecified three-layer set of contrasts: layer 1, NPM vs. HC; layer 2, IBC vs. IDC; and layer 3, IBC vs. NPM. For each layer, multivariable logistic regression estimated the adjusted odds ratio (OR) for log-transformed PRL, with adjustment for age, BMI, and menopausal status. Discrimination was quantified as the area under the receiver operating characteristic curve (AUC) for PRL alone and for the covariate-adjusted model.

Combined model and robustness

To test whether PRL adds discriminative value beyond systemic inflammation for separating IBC from NPM, we fitted a four-variable logistic model specified a priori, including log(PRL), log(CRP), log(NLR), and age. Performance was assessed according to the apparent AUC with bootstrap internal validation for the optimism-corrected AUC; the calibration curve was generated with 200 bootstrap resamples. Calibration was additionally assessed by the Hosmer-Lemeshow test, and the combined model was compared against single markers using via DeLong test for correlated ROC curves (28). An exploratory nomogram and decision-curve analysis (29) were generated as reference analyses. Because several IBC-related models had modest events per variable (≈10), all key logistic models were re-estimated with Firth penalized maximum-likelihood regression (30,31).

Correlation and IBC-specific analyses

Within-group associations between PRL and inflammatory markers, cytokines, and pathological variables were assessed via Spearman rank correlation (ρ); within-IBC and within-NPM cytokine correlations were adjusted for multiple comparisons via the Benjamini-Hochberg false-discovery rate (FDR), with q values being reported (32), and between-group differences in correlation strength were tested by Fisher r-to-z transformation. Associations between PRL and pretreatment tumor burden in IBC were examined by Spearman correlation, a median-split Mann-Whitney U comparison, and the locally estimated scatterplot smoothing (LOESS) curve, with a prespecified robustness check excluding 13 patients with IBC and PRL >496 mIU/L. A multivariable linear regression model with log(PRL) as the dependent variable [covariates: log(ultrasound diameter), log(NLR), age, and BMI] was fitted and reported descriptively. The distribution of hyperprolactinemia across the five molecular subtypes was compared via a Monte Carlo permutation chi-square test (10,000 permutations), and a post hoc dichotomous contrast (luminal B HER2+ plus TNBC versus the remaining subtypes), defined after inspection of the subtype distribution and reported as exploratory, was evaluated with the Fisher exact test.

Sensitivity analyses and general considerations

Menopausal status-stratified analyses were performed for the IBC-versus-IDC contrast. Because only one patient with NPM was postmenopausal, contrasts involving NPM could not be meaningfully stratified and are reported as overall estimates adjusted for age and BMI. Effect modification in the IBC-versus-IDC contrast was assessed using a log(PRL)-by-menopausal-status interaction term. A further robustness analysis excluded 11 IBC cases with a disease course exceeding 180 days. All tests were two-sided, with P<0.05 being considered significant. Given the small IBC sample (n=42), nonsignificant findings are interpreted as an absence of detectable difference rather than proof of equivalence, and the IBC-specific analyses are reported as exploratory. Analyses were performed with standard statistical software for the corresponding methods.

Handling of PRL confounders and sampling considerations

Because serum PRL is sensitive to physiological, pharmacological, and pathological influences, confounding was addressed through a staged strategy. At the design stage, the prespecified PRL-confounding conditions and medications listed in section “Diagnostic criteria and group definitions” were treated as exclusion criteria. At the analysis stage, between-group contrasts were adjusted for age, BMI, and menopausal status, and the IBC-versus-IDC comparison was additionally examined within menopausal strata.

Two features of routine PRL measurement could not be standardized retrospectively. Blood sampling was not fixed to a morning time window, and macroprolactin was not routinely assayed across the cohort. The central laboratory reference range was therefore used consistently, with the upper limit of normal for nonpregnant women (496 mIU/L) indicating hyperprolactinemia a priori.

Results

Study population and baseline characteristics

A total of 512 women were included in the final analysis, comprising 121 HCs, 123 patients with IDC, 42 patients with IBC, and 226 patients with NPM. The four groups differed significantly in baseline demographics and peripheral blood markers (Kruskal-Wallis P<0.001 for all continuous variables; chi-square P<0.001 for menopausal status). HCs and patients with NPM were comparatively young (median age 34 and 34.5 years, respectively) and were predominantly premenopausal (95.0% and 99.6%, respectively). The IDC group was the oldest (median age 61 years; 30.9% premenopausal), followed by the IBC group (median age 53.5 years; 47.6% premenopausal). The median BMI ranged narrowly from 21.5 to 22.9 kg/m². Detailed baseline characteristics are summarized in Table 1.

Table 1. Baseline demographic and peripheral blood characteristics across the four groups.

Variable	HC (n=121)	IDC (n=123)	IBC (n=42)	NPM (n=226)	P value
Age (years)	34 (30–39)	61 (46–68)	54 (38–68)	34 (32–38)	<0.001
BMI (kg/m²)	21.5 (20.2–23.4)	22.5 (21.5–24.0)	22.9 (21.6–24.2)	22.4 (20.2–23.9)	<0.001
Premenopausal	115 (95.0)	38 (30.9)	20 (47.6)	225 (99.6)	<0.001
PRL (mIU/L)	303 (237–370)	208 (148–286)	342 (228–707)	492 (280–781)	<0.001
CRP (mg/L)	0.52 (0.29–1.19)	0.50 (0.50–1.58)	2.36 (0.50–5.37)	7.51 (2.21–23.58)	<0.001
WBC (×10⁹/L)	5.80 (4.70–7.10)	5.70 (5.00–7.20)	6.25 (5.33–7.50)	8.55 (7.00–10.17)	<0.001
Neutrophil count (×10⁹/L)	3.60 (2.90–4.80)	3.66 (2.85–4.60)	4.21 (3.35–5.24)	6.11 (4.80–7.79)	<0.001
NLR	2.29 (1.67–3.00)	2.32 (1.61–3.08)	2.80 (2.00–3.58)	3.71 (2.60–5.03)	<0.001
SII	553 (386–725)	533 (363–814)	745 (433–1,155)	1,176 (727–1,645)	<0.001

Values are median (IQR) for continuous variables and n (%) for categorical variables. P values are from the Kruskal-Wallis test (continuous variables) and chi-square test (menopausal status). Menopausal status was harmonized from group-specific source codes; codes 1–3 versus 4 in HC and NPM and codes 1 versus 0 in IDC and IBC; these denote premenopausal and postmenopausal status, respectively. For IDC, BMI was computed from recorded height and weight. BMI, body mass index; CRP, C-reactive protein; HC, healthy control; IBC, inflammatory breast cancer; IDC, invasive ductal carcinoma; IQR, interquartile range; NLR, neutrophil-to-lymphocyte ratio; NPM, nonpuerperal mastitis; PRL, prolactin; SII, systemic immune-inflammation index; WBC, white blood cell count.

Group-wise distribution of inflammatory-endocrine markers

The four-group distribution of PRL and inflammatory markers is shown in Figure 2. PRL displayed a nonmonotonic pattern across diagnostic categories: median PRL was highest in the NPM group (492 mIU/L), followed by the IBC (342 mIU/L) and HC (303 mIU/L) groups, with IDC showing the lowest values (208 mIU/L). In contrast, conventional inflammatory markers (CRP, WBC, neutrophil count, NLR, and SII) followed a strictly inflammation-graded pattern, with NPM consistently being highest, IBC intermediate, and HC and IDC the lowest. The age distribution clearly separated the two younger groups (HC and NPM) from the two older groups (IDC and IBC). NPM exhibited the most pronounced acute-inflammation phenotype across all conventional markers, while IBC displayed an intermediate inflammatory profile but a PRL level comparable to that of NPM.

Figure 2.

Box plots of key inflammatory and endocrine markers across the four diagnostic groups (HC, IDC, IBC, and NPM). (A-G) PRL, CRP, WBC, neutrophil count, NLR, PLR, and SII (PRL, CRP, and SII are plotted on log10 y-axes; the remaining panels include linear scales). (H) The age distribution. NPM has the highest values across all inflammatory markers, IBC occupies an intermediate position, and IDC and HC remain close to baseline; PRL is highest in NPM and IBC and lowest in IDC. CRP, C-reactive protein; HC, healthy control; IBC, inflammatory breast cancer; IDC, invasive ductal carcinoma; NLR, neutrophil-lymphocyte ratio; NPM, nonpuerperal mastitis; PLR, platelet-to-lymphocyte ratio; PRL, prolactin; SII, systemic immune-inflammation index; WBC, white blood cell count.

To visualize this multidimensional phenotype landscape, we plotted serum PRL against CRP in two-dimensional log-log space and performed an exploratory PCA of seven log-transformed inflammatory-endocrine markers and age (Figure S1). The PRL-CRP phenotype map showed that NPM was shifted toward the high-PRL/high-CRP region, IBC formed an intermediate mixed cluster overlapping with both NPM and IDC, and IDC and HC formed two largely overlapping low-CRP baseline clusters. PCA revealed distinct group-centroid locations: PC1 (55.3% of total variance) loaded heavily on log(PRL) and the inflammatory markers, while PC2 (15.6%) was driven primarily by age. Substantial overlap nevertheless remained in the individual-level distributions and confidence ellipses.

Three-layer comparison of PRL across diagnostic contrasts

To dissect the diagnostic information carried by PRL, we performed a prespecified three-layer comparison using multivariable logistic regression adjusted for age, BMI, and menopausal status (Table 2 and Figure 3A).

Table 2. Three-layer comparison of PRL across the NPM vs. HC, IBC vs. IDC, and IBC vs. NPM .

Comparison	Case/control, n	PRL (mIU/L), median (IQR)	Adj. OR for log(PRL) (95% CI)	P	AUC PRL	AUC combined
NPM vs. HC	226/121	492 (280–781) vs. 303 (237–370)	3.88 (2.52–6.20)	<0.001	0.717	0.739
IBC vs. IDC	42/123	342 (228–707) vs. 208 (148–286)	4.03 (2.19–7.97)	<0.001	0.727	0.724
IBC vs. NPM	42/226	342 (228–707) vs. 492 (280–781)	0.94 (0.31–2.97)	0.91	0.411	0.895

Adjusted ORs from multivariable logistic regression with log(PRL), adjusted for age, BMI, and menopausal status. The AUC of PRL alone for the IBC-vs-NPM comparison is below 0.5 because serum PRL is lower in IBC than in NPM; the corresponding discriminative AUC is 0.589. AUC, area under the receiver operating characteristic curve; BMI, body mass index; CI, confidence interval; HC, healthy control; IBC, inflammatory breast cancer; IDC, invasive ductal carcinoma; IQR, interquartile range; NPM, nonpuerperal mastitis; OR, odds ratio; PRL, prolactin.

Figure 3.

Three-layer comparison of inflammatory-endocrine markers and within-group correlations with PRL. (A) Forest plot of adjusted odds ratios per one-unit increase in log-transformed PRL, CRP, NLR, and SII for the three-layered comparisons (NPM vs. HC, IBC vs. IDC, and IBC vs. NPM) from multivariable logistic regression adjusted for age, BMI, and menopausal status. (B) Heat map of within-group Spearman correlation coefficients between PRL and inflammatory markers (CRP, NLR, PLR, MLR, SII, and WBC) in NPM, IBC, and IDC. *P<0.05; **P<0.01; ***P<0.001. BMI, body mass index; CI, confidence interval; CRP, C-reactive protein; HC, healthy control; IBC, inflammatory breast cancer; IDC, invasive ductal carcinoma; MLR, monocyte-lymphocyte ratio; NLR, neutrophil-lymphocyte ratio; NPM, nonpuerperal mastitis; OR, odds ratio; PLR, platelet-to-lymphocyte ratio; PRL, prolactin; SII, systemic immune-inflammation index; WBC, white blood cell count.

Layer 1 (NPM vs. HC)

In this age-matched comparison (median age 34 vs. 34 years), log(PRL) was strongly associated with NPM (adjusted OR 3.88; 95% CI: 2.52–6.20; P<0.001), with an apparent AUC of 0.717 for PRL alone and 0.739 for the combined model. CRP, NLR, and SII were also strongly associated with NPM in the same direction.

Layer 2 (IBC vs. IDC)

Restricted to malignant tumors, log(PRL) remained strongly associated with the inflammatory cancer phenotype (adjusted OR 4.03; 95% CI: 2.19–7.97; P<0.001), with an AUC of 0.727 for PRL alone and 0.724 for the combined model.

Layer 3 (IBC vs. NPM)

In the clinically critical contrast between the two “inflammatory” mimickers, log(PRL) showed no independent discriminative value (adjusted OR 0.94; 95% CI: 0.31–2.97; P=0.91; discriminative AUC 0.589). Given the IBC sample size (n=42), this nonsignificant result indicates an absence of detectable difference rather than demonstrated equivalence. None of the conventional inflammatory markers significantly distinguished the two conditions after adjustment [log(CRP): OR 0.81, 95% CI: 0.41–1.55, P=0.52; log(NLR): OR 0.45, P=0.31; log(SII): OR 0.72, P=0.58], although every inflammatory-marker point estimate lay below 1, consistent with the substantially higher absolute CRP and inflammatory burden in NPM than in IBC. A multivariable combined model nonetheless achieved high discrimination (apparent AUC 0.895; see section “Combined modelling of PRL and inflammatory markers for the IBC-versus-NPM contrast”).

Combined modelling of PRL and inflammatory markers for the IBC-versus-NPM contrast

Because PRL alone did not discriminate IBC from NPM, we sought to determine whether PRL adds discriminative value when combined with inflammatory markers. We fitted a multivariable logistic model integrating log(PRL), log(CRP), log(NLR), and age, specified a priori. Single markers were weak discriminators, with the highest discriminative AUC being 0.752 for WBC (and the lowest being 0.589 for PRL), while the combined model achieved a higher apparent AUC of 0.895. However, the contribution of PRL was null: in the combined model, the only independent predictors were log(CRP) (OR 0.68; 95% CI: 0.48–0.98; P=0.04) and age (OR 1.19; 95% CI: 1.13–1.25; P<0.001) and not log(PRL) (OR 1.19; P=0.62) or log(NLR) (OR 1.08; P=0.89). The model’s discrimination is therefore carried entirely by systemic inflammatory burden and age, not by PRL.

The CRP effect was inverse (higher CRP lowering the odds of IBC) because absolute CRP was substantially higher in NPM than in IBC. The full model coefficients, internal validation (bootstrap optimism correction, DeLong comparisons against single markers, Hosmer-Lemeshow calibration, and Firth penalized re-estimation), and an exploratory nomogram with DCA are provided in Table S1 and Figure S2.

Relationship between PRL and inflammation: from upstream cytokines to peripheral indices

Upstream cytokines

Serum IL-6, IL-1β, and TNF-α were available for all HCs and patients with NPM and for subsets of IBC (n=31) and IDC (n=56). All three cytokines differed across the four groups (Kruskal-Wallis P<0.001), with the highest level of IL-6 and IL-1β in NPM. Within the NPM group (n=226), PRL was significantly and positively correlated with all three cytokines (IL-6: ρ=0.34; IL-1β: ρ=0.25; TNF-α: ρ=0.26; all P values <0.001; FDR q<0.001). In IBC, no such coupling was detectable (ρ≈0.03–0.06; P>0.77), and no significant coupling was found in the IDC or HC groups. Fisher r-to-z comparison of the NPM and IBC coefficients did not reach significance (IL-6: P=0.10; IL-1β: P=0.264; TNF-α: P=0.30), and thus the apparent between-group difference could not be confirmed given the limited IBC cytokine subset (Table S2).

Peripheral inflammatory indices

When the PRL-inflammation correlation analysis was repeated using conventional peripheral inflammatory indices, the results were qualitatively similar to those of the cytokine analysis: IDC consistently showed no positive PRL-inflammation coupling, with several markers showing weak negative correlations (ρ=−0.18 to 0.12)—but differed for the inflammatory conditions. Positive coupling was found for both IBC (ρ=0.24–0.42; all markers except PLR were significant after FDR adjustment) and NPM (ρ=0.06–0.31), and the between-group difference was not statistically significant for any marker (all Fisher r-to-z transformation P≥0.12; Table S3 and Figure 3B). Across both analyses, one feature was consistent: IDC was decoupled, while at least one of the inflammatory conditions demonstrated PRL-inflammation coupling.

Association of PRL with pretreatment tumor burden in IBC (Figure 4)

Figure 4.

Association of serum PRL with pretreatment tumor burden in IBC. (A) Scatter plot of log10-transformed serum PRL versus pretreatment ultrasound maximum tumor diameter (mm), with a LOESS curve and points coded by menopausal status (Spearman ρ=0.46; P=0.002; n=42). (B) Box plot of serum PRL stratified by maximum ultrasound-derived tumor diameter using a 50-mm cutoff (50 mm): the median PRL was 248 mIU/L for a diameter <50 mm and 480 mIU/L for a diameter ≥50 mm (Mann-Whitney U P=0.009). IBC, inflammatory breast cancer; LOESS, locally estimated scatterplot smoothing; PRL, prolactin.

Because most patients with IBC received neoadjuvant therapy that markedly reduced lesion size before surgery, pretreatment ultrasound and MRI measurements were used as the burden indicator rather than posttreatment pathological dimensions. Within the IBC group, serum PRL was positively correlated with the pretreatment maximum tumor diameter measured by ultrasound (n=42; ρ=0.46; P=0.002), which was corroborated by an MRI-examined subset (n=28; ρ=0.49; P=0.008). Using a 50-mm cutoff, patients with a maximum ultrasound-derived tumor diameter ≥50 mm (n=22) had a higher median serum PRL level than those with a diameter <50 mm (n=20) (480 vs. 248 mIU/L, respectively; Mann-Whitney U P=0.009; Figure 4B). A LOESS curve indicated that PRL remained near baseline for tumors below approximately 50 mm and rose more steeply thereafter (Figure 4A).

Sensitivity to high-PRL cases

In a prespecified robustness check, excluding the 13 patients with IBC with PRL >496 mIU/L, 29 patients remained for the ultrasound diameter analysis; the correlation between PRL and ultrasound tumor diameter to nonsignificance (n=29; ρ=0.17; P=0.37). The PRL-tumor-burden association should therefore be interpreted as being driven primarily by a high-PRL subgroup rather than as a continuous linear dose-response relationship.

A multivariable linear regression with log(PRL) as the dependent variable indicated that pretreatment ultrasound tumor size and systemic inflammation each contributed to circulating PRL (Table 3). After mutual adjustment, log(PRL) was associated with log(ultrasound diameter) (β=0.67; 95% CI: 0.10–1.23; P=0.02) and log(NLR) (β=0.67; 95% CI: 0.22–1.12; P=0.005) but not with age or BMI (P>0.8). The model explained 34% of the variance (R²=0.34; adjusted R²=0.27; F test P=0.003) and should be regarded as descriptive rather than predictive.

Table 3. Association of PRL with pretreatment tumor burden and inflammatory markers in patients with IBC.

Panel	Variable	n	Statistic	Value	SE	95% CI	P value
A: Spearman correlation	Pretreatment max. tumor diameter on ultrasound (mm)	42	ρ	0.46	—	—	0.002
	Pretreatment max. tumor diameter on MRI (mm)	28	ρ	0.49	—	—	0.008
	Ultrasound diameter after exclusion of patients with PRL >496 mIU/L	29	ρ	0.17	—	—	0.37
B: multivariable linear regression for log(PRL)	Log(max diameter on ultrasound)	–	β	0.67	0.28	0.10 to 1.23	0.02
	Log(NLR)	–	β	0.67	0.22	0.22 to 1.12	0.005
	Age	–	β	−0.002	0.007	−0.02 to 0.01	0.82
	BMI	–	β	−0.009	0.043	−0.10 to 0.08	0.84

Panel A: Spearman correlation analysis. Panel B: multivariable linear regression with log(PRL) as the dependent variable; model R²=0.34, adjusted R²=0.27, F test P=0.003. β, unstandardized regression coefficient; BMI, body mass index; CI, confidence interval; IBC, inflammatory breast cancer; MRI, magnetic resonance imaging; NLR, neutrophil-to-lymphocyte ratio; ρ, Spearman rank correlation coefficient; PRL, prolactin; SE, standard error.

Exploratory pathological correlations

Within the IBC group, serum PRL was not significantly correlated with the Ki-67 index (ρ=0.01; P=0.94), ER% (ρ=−0.27; P=0.08), or PR% (ρ=−0.23; P=0.15). Lymph node status was not examined because the available nodal counts reflected postneoadjuvant (yp) staging. These analyses were exploratory and underpowered.

Distribution of hyperprolactinemia across molecular subtypes within IBC

Although serum PRL did not distinguish IBC from NPM, we sought to determine whether the magnitude of PRL elevation carries information within IBC itself. It was found that 13 of 42 patients with IBC (31%) met the prespecified criterion for hyperprolactinemia, defined as serum PRL >496 mIU/L (the clinical laboratory upper limit of normal for nonpregnant women). The distribution of hyperprolactinemia across the five molecular subtypes was nonuniform (Monte Carlo permutation P=0.03; Table 4). Hyperprolactinemia appeared concentrated in luminal B HER2+ (6/10, 60%) and TNBC (5/11, 45%) tumors, occurred in 2 of 9 luminal B HER2− cases (22%), and was not observed in luminal A (0/8) or HER2-enriched (0/4) tumors. These subtype-specific rates were based on small per-subtype counts and had wide CIs; in particular, the 0% rates in luminal A and HER2-enriched tumors (n=8 and n=4) had upper 95% confidence limits of approximately 37% and 60%, respectively, and were therefore statistically compatible with the 31% overall rate; thus, the apparent absence in these subtypes should be regarded as hypothesis-generating rather than established. A post hoc dichotomous comparison contrasting the two most aggressive subtypes (luminal B HER2+ and TNBC) against the remaining three—an aggregation defined after the subtype distribution had been examined, with the resulting OR and P value being descriptive and likely optimistic—nonetheless showed a strong association: hyperprolactinemia was present in 11 of 21 patients in the aggressive group but only in 2 of 21 in the remainder group (Fisher exact OR 10.4; 95% CI: 1.9–56.7; P=0.006). As an exploratory indicator for these aggressive subtypes within IBC, hyperprolactinemia showed a sensitivity of 0.52, a specificity of 0.90, a positive predictive value of 0.85, and a negative predictive value of 0.66; as these operating characteristics depend on the subtype mix and IBC prevalence of this sample, they are reported descriptively and are not intended to be applied in other settings. Patients with hyperprolactinemia also had larger median pretreatment tumor diameters on ultrasound (60 vs. 46 mm; P=0.004) (Table S4).

Table 4. Distribution of hyperprolactinemia (serum PRL >496 mIU/L) across the molecular subtypes of IBC.

Molecular subtype	n	Hyperprolactinemia, n (% of all hyperPRL)	Normoprolactinemia, n (% of all normoPRL)	Within-subtype rate
Luminal A	8	0 (0)	8 (28)	0/8 (0%)
Luminal B HER2−	9	2 (15)	7 (24)	2/9 (22%)
Luminal B HER2+	10	6 (46)	4 (14)	6/10 (60%)
HER2-enriched	4	0 (0)	4 (14)	0/4 (0%)
TNBC	11	5 (38)	6 (21)	5/11 (45%)
Total	42	13 (100)	29 (100)	13/42 (31%)

Hyperprolactinemia was defined as serum PRL >496 mIU/L. Molecular subtypes were inferred from immunohistochemistry surrogates (ER, PR, HER2, and Ki-67) according to the St. Gallen criteria; HER2 positivity was defined as IHC 3+ or IHC 2+ with FISH amplification. Monte Carlo permutation chi-square test: χ²=10.73 and P=0.026 (10,000 permutations). Post hoc dichotomous comparison of luminal B HER2+ plus TNBC versus the other three subtypes (defined after the distribution was inspected and considered exploratory): Fisher exact OR =10.4 (95% CI: 1.9–56.7) and P=0.006. ER, estrogen receptor; FISH, fluorescence in situ hybridization; HER2, human epidermal growth factor receptor 2; hyperPRL, hyperprolactinemia; IBC, inflammatory breast cancer; IHC, immunohistochemistry; normoPRL, normoprolactinemia; OR, odds ratio; PR, progesterone receptor; PRL, prolactin; TNBC, triple-negative breast cancer.

Sensitivity analyses

Menopausal-status stratification

An age-stratified sensitivity analysis of the three-layer comparisons and the IBC-versus-NPM combined model is presented in Figure S3. Because only one patient with NPM was postmenopausal (pre-menopausal: 225/226, 99.6%), the comparison involving NPM could not be meaningfully stratified by menopausal status. Overall estimates adjusted for age and BMI were therefore reported for layer 1 (NPM vs. HC: OR 4.18; 95% CI: 2.73–6.65; P<0.001; n=347) and layer 3 (IBC vs. NPM: OR 0.91; 95% CI: 0.48–1.69; P=0.76; n=268). The IBC-versus-IDC comparison was stratified by menopausal status. Among premenopausal women (IBC: n=20; IDC: n=38), the adjusted OR for log(PRL) was 2.03 (95% CI: 0.91–5.10; P=0.10; n=58). Among postmenopausal women (IBC: n=22; IDC: n=85), the corresponding OR was 7.63 (95% CI: 3.02–23.79; P<0.001; n=107). The difference between strata did not reach conventional statistical significance [log(PRL)-by-menopausal-status interaction P=0.055]. These findings suggest possible effect modification by menopausal status but should be interpreted cautiously given the limited IBC sample size. The combined IBC-versus-NPM model retained high discrimination overall (apparent AUC 0.895; section “Combined modelling of PRL and inflammatory markers for the IBC-versus-NPM contrast”); stratum-specific AUCs are not reported because the near-absence of postmenopausal NPM renders them uninformative.

Robustness checks

Excluding 11 IBC cases with a disease course beyond 180 days (with 31 cases with a more typical acute presentation being retained) strengthened the layer 2 association (adjusted OR 6.02; P<0.001) and preserved the layer 3 null (OR 1.18; P=0.64), with the combined model retaining high discrimination (Table S5). Because the IBC-related models had modest events per variable (≈10), all key models were also refitted with Firth penalized regression; the estimates were concordant throughout (Table S1).

Summary of key findings

In summary, the results support a two-level interpretation. Across diagnostic groups, PRL served as a shared inflammatory-state marker: it was elevated in NPM and IBC relative to their noninflammatory comparators, but it did not distinguish IBC from NPM. Within IBC, however, the magnitude of PRL elevation carried additional exploratory information because hyperprolactinemia was more common in luminal B HER2+ and TNBC tumors and was linked to the larger-tumor, high-PRL subgroup. These findings motivated the mechanistic and translational interpretation outlined below.

Discussion

This study provides, to our knowledge, the first systematic characterization of serum PRL in IBC, together with the first direct comparison of PRL between patients with IBC and those with NPM. Three principal findings emerged from this study. First, PRL was elevated in both inflammatory conditions relative to their respective comparators: the median level in the NPM group exceeded that in the HC group (adjusted OR 3.88), while that in the IBC exceeded that of the IDC group (adjusted OR 4.03; 342 vs. 208 mIU/L). Second, PRL did not differ between the IBC and NPM (adjusted OR 0.94; P=0.91). Third, PRL was positively associated with systemic inflammatory markers in both the IBC and NPM groups but not with the IDC groups. Overall, this pattern indicates that PRL elevation, rather than indicating a specifying, diagnosis, marks an inflammation-associated endocrine state shared by inflammatory breast diseases (33,34).

The findings from the IBC-versus-NPM comparison were consistent with our prespecified expectation that PRL would not distinguish the two inflammatory conditions. PRL alone was a weak discriminator, and in the four-variable model, its contribution was not independent; the apparent separation of the two conditions was instead driven by CRP and age. This pattern is biologically coherent: a marker comparably elevated in both inflammatory conditions should not be expected to distinguish them. Although this can be construed as a negative technical result, it also supports PRL as being part of a shared inflammation-associated endocrine state. A breast-specific basis for this link has also been proposed: in patients with granulomatous and NPM, PRL is thought to drive the local inflammatory response through NF-κB-mediated induction of pro-inflammatory cytokines (IL-1, IL-6, and TNF-α) in the mammary epithelium, and reports of mastitis resolving after correction of hyperprolactinemia also attest to the contributory role of PRL (7-9). The inverse CRP effect is also clinically useful from a different perspective: among patients with an inflammation of the breast, a relatively modest CRP should not be interpreted as reassuring because the more florid systemic inflammatory response was found to be characteristic of NPM and not IBC.

If PRL is a shared inflammation-associated signal, the question arises as to which layer of inflammation drives it. A direct examination of upstream inflammatory cytokines sharpened—and partly reframed—the mechanistic interpretation. In NPM, serum PRL was significantly and positively correlated with IL-6, IL-1β, and TNF-α levels (P<0.001; FDR q<0.001), providing internally derived support for the proposed cytokine-to-PRL link of the IPP axis in a benign inflammatory condition (10,12,23). In IBC, no such cytokine coupling was detectable, although the between-group difference did not reach statistical significance in the small IBC cytokine subset (Fisher r-to-z transformation for IL-6: P=0.10). Conventional peripheral inflammatory indices suggested a somewhat different pattern: both inflammatory conditions showed positive PRL-index coupling, while IDC remained decoupled. This discrepancy is itself informative: peripheral indices appear to capture a broader, downstream perspective of systemic inflammation, whereas upstream cytokines may capture a more proximal, mechanism-specific signal—to which IBC may not respond in the same manner as NPM. Such a divergence is consistent with the increasingly recognized view that, despite its overtly inflammatory clinical presentation, IBC is underpinned by a biologically distinct and largely immunosuppressive tumor microenvironment, whose cellular composition and cytokine wiring differ markedly from that of benign inflammatory breast disease (35-38). Notably, the tumor-associated macrophages of IBC have been shown to secrete a chemotactic cytokine program (including TNF-α and IL-8) that drives the motility and invasion of carcinoma cells, illustrating how the deployment of the same cytokines may differ across the IBC microenvironment and benign inflammation (38). Peripheral blood inflammatory ratios such as the NLR and PLR have previously been shown to carry prognostic and biological information in patients with IBC and to reflect distinctive, partly suppressed immune microenvironment of this condition, supporting their use here as a meaningful axis of systemic inflammation (39). Together, the findings suggest a more refined hypothesis: PRL is elevated in both inflammatory conditions, but this increase is only cytokine-coupled in NPM; indeed, based on underpowered data, this elevation was not linked with cytokine level in IBC—a between-group difference that did not itself reach statistical significance. If confirmed in larger cohorts, this would suggest that two clinically similar conditions exhibit a comparable PRL phenotype through partly different mechanisms: cytokine-driven pituitary stimulation in benign inflammation and possibly a tumor-intrinsic mechanism in IBC.

Perhaps the most generalizable finding in this study concerns a subset of IBC patients with hyperprolactinemia. Defined a priori by the clinical laboratory upper limit of normal (PRL >496 mIU/L) rather than a data-derived cutoff, this hyperprolactinemic subset included 13 of 42 patients with IBC (31%) and appeared to have a high proportion of the two most aggressive molecular subtypes—luminal B (HER2+) and TNBC—In a post hoc dichotomous comparison, we combined luminal B (HER2+) and TNBC into one group and compared it with the remaining three subtypes (luminal A, luminal B HER2−, and HER2-enriched). This comparison—an aggregation selected after the subtype distribution had been determined—reached nominal significance (Fisher OR 10.4; 95% CI: 1.9–56.7; P=0.006). No cases with luminal A or HER2-enriched tumors had hyperprolactinemia; however, with only 8 and 4 patients in these subtypes, respectively, the observed 0% rates carry wide CIs (exact upper 95% limits approximately 37% and 60%, respectively) and remain statistically compatible with the 31% overall rate; thus, this apparent absence must be interpreted cautiously. This entire subtype analysis is hypothesis-generating only, and prospective validation in an independent cohort is required before any subtype-specific conclusion can be drawn. The associated predictive values—specifically, the specificity of 0.90 and the positive predictive value of 0.85—are likewise contingent on the subtype mix and IBC prevalence of this single-center sample and not generalizable. Given the well-documented aggressiveness and poor prognosis associated with IBC (1,40) and consistent with earlier observations that elevated serum PRL predicts poorer endocrine-therapy response and shorter survival in patients with advanced breast cancer (21), we believe that further validation of hyperprolactinemia as a candidate biological marker among patients with IBC is warranted. In this sample, a single routine serum measurement identified, with high specificity (90%), a subset with a higher frequency of the subtypes that drive the disease’s aggressive course. Because the molecular subtype can itself be determined on the diagnostic biopsy, the value of this signal is unlikely to lie in pre-empting subtyping: it is better regarded as a biological and mechanistic lead—and a candidate marker for PRLR-directed translational study—than as a stand-alone triage test.

Biologically, the distribution of hyperprolactinemia in luminal B HER2+ and TNBC tumors is coherent with the known overlap between PRL/PRLR signaling and the growth-signaling networks that are highly active in these subtypes, rather than being incidental. Luminal B HER2+ and TNBC tumors converge on highly active growth-signaling networks—HER2/PI3K-AKT and MAPK for the former and a heterogeneous mix of receptor tyrosine kinase, PI3K-AKT, and MAPK activation for the latter. PRL signaling, mediated by the PRLR, engages exactly the same downstream cascades: JAK2-STAT5, PI3K-AKT, and MAPK (15,17,41,42). Indeed, in ER+ breast cancer cells, PRL, acting through PRLR, has been shown to drive transcriptional upregulation of PRLR itself through STAT5 and ERα cross-talk and to induce JAK2-dependent HER2 phosphorylation, independently of estradiol (17). This constitutes a mechanistic basis for an oncogenic PRL loop that overlaps with the growth-signaling program of these aggressive subtypes. Conversely, CRISPR-mediated loss of PRLR in luminal (MCF-7) and HER2-enriched (SKBR-3) breast cancer cells promotes dedifferentiation, stem-like plasticity, upregulation of HER2, enhanced tumorigenic and metastatic capacity, and resistance to antihormonal and anti-HER2 therapy. In our study, hyperprolactinemia was associated with aggressive, growth-signaling-active molecular subtypes (luminal B HER2+ and TNBC). This observation is consistent with evidence that the balance of PRL/PRLR signaling can shape such aggressive phenotypes (42). The concentration of hyperprolactinemia in these IBC subtypes, whose growth signaling overlaps with PRL’s downstream effects, and its absence from cases with luminal A (low proliferation) and pure HER2-enriched tumors, is consistent with PRL/PRLR signaling acting as an auxiliary growth-supporting pathway in growth-signaling-active IBC. This in turn raises speculation that may have a tangible translational implication: the hyperprolactinemic, aggressive-subtype-enriched IBC subset may be susceptible to PRLR-directed therapy. Pharmacological PRLR antagonism (e.g., the G129R analog or the LFA102 antibody) has been investigated preclinically in breast cancer (43), and although a phase I trial of LFA102 monotherapy did not demonstrate clinical responsiveness in unselected patients with advanced breast and prostate cancer (44), it has not been tested in a biomarker-selected, IBC-specific context. Our data identify a candidate population for which such an approach may be biologically plausible. However, this interpretation remains indirect because we measured circulating PRL, not tumor PRLR expression or PRL/PRLR signaling activation. Subtype-specific PRLR data are heterogeneous: PRLR protein is highest in luminal A and low/absent in TNBC in some series (45), and PRLR mRNA is decreased in HER2-positive ER−, PR− tumors, with proposed subtype-dependent oncogenic versus tumor-suppressive roles (46); conversely, a PRL pro-differentiation pathway associated with better prognosis has been described in TNBC (47). Thus, hyperprolactinemia distribution in luminal B HER2+ and TNBC cannot be equated with PRL/PRLR pathway activation; if these tumors are relatively PRLR-low, circulating PRL elevation may reflect systemic inflammation rather than tumor pathway engagement. The therapeutic speculation regarding PRLR-directed therapy should therefore be understood as an indirect, hypothesis-generating inference contingent on future tissue-based demonstration of PRLR expression and signaling in IBC.

Other findings generated in our study are hypothesis-generating and merit prospective study. The lack of association between PRL and inflammatory cytokines in the IBC groups, contrasted with their strong correlation in the NPM groups, suggests that PRL elevation in IBC may have a tumor-intrinsic, cytokine-independent origin. A cytokine-independent, locally produced source of PRL would be consistent with the established capacity of nonpituitary cells, including leukocytes, fibroblasts, and epithelial cells, to synthesize PRL and signal through PRLR in an autocrine or paracrine manner (11,13,14,17). Alternatively, the immunosuppressive cytokine network of the IBC microenvironment may uncouple systemic cytokine levels from PRL release, such that PRL elevation in IBC reflects a different inflammatory axis than in NPM. These possibilities are not mutually exclusive and require tissue-level studies of PRL expression, PRLR status, and local cytokine profiles. This finding, together with the subtype distribution described above, is consistent with previous reports of tumor PRLR expression and paracrine/autocrine PRL signaling in breast cancer (24,41) and supports the rationale for confirmatory tissue-based studies of PRLR expression and local PRL production in IBC; until such data are available, the link between serum PRL and tumor PRL/PRLR signaling remains indirect. The correlation between PRL and pretreatment tumor burden in IBC (ρ=0.46) was reduced to nonsignificance after 13 patients with PRL >496 mIU/L were excluded (ρ=0.17; P=0.37), indicating that the apparent linear association was driven by the hyperprolactinemic subgroup rather than by a continuous dose-response across the IBC range. Inflammatory breast conditions may, more generally, be more naturally stratified along a two-dimensional “PRL × inflammatory intensity” phenotype space than along any single marker.

Clinically, these findings should be read as foundational rather than diagnostic. The study addresses a deficiency in the literature by characterizing serum PRL in IBC and locating it within a four-state comparison of healthy, benign-inflammatory, noninflammatory-malignant, and inflammatory-malignant breast conditions. The IBC-versus-NPM combined model is exploratory and may generate hypotheses for further investigation, but it is not intended for clinical triage and should not be regarded as a substitute for biopsy or as a stand-alone diagnostic test. The principal contribution of this study is therefore conceptual: PRL is similarly elevated in NPM and IBC and does not distinguish between them, whereas systemic inflammation (CRP) and patient age account for the separation between the two conditions observed in this cohort. Future discovery work using banked blood samples from this cohort may help identify additional biomarkers of IBC and is warranted.

Limitations

Design and causality

This was a single-center, retrospective, case-control study without external validation. The associations are cross-sectional and cannot establish whether inflammation drives PRL elevation, whether PRL modulates inflammation, or whether both reflect a common upstream process; the IPP axis is proposed as a framework rather than as a demonstrated causal mechanism.

Sample size and precision

The IBC group comprised only 42 patients, and the IBC-related models had modest events per variable (≈10). Although Firth penalized regression produced concordant estimates, the within-group correlations and the tumor-burden association in IBC should be regarded as exploratory and require replication. Cytokine measurements were available for only a subset of the IBC (n=31) and IDC (n=56) groups, and thus the absence of a detectable PRL-cytokine correlation in IBC may partly reflect limited power. Moreover, the apparent NPM-versus-IBC difference in cytokine coupling did not reach statistical significance, and the menopausal-status-stratified analyses were limited by the small number of patients with IBC within each stratum and by the near absence of postmenopausal patients in the NPM group.

PRL is not an independent discriminator

In the combined IBC-versus-NPM model, discrimination was driven by CRP and age; PRL and NLR were not independently significant and were retained only as concept variables. The model should not be construed as a PRL-based diagnostic instrument.

Measurement and definitional considerations

Peripheral blood PRL does not necessarily reflect the local tumor microenvironment, and we did not measure tumor PRLR expression, PRL/PRLR signaling activation, or local PRL production; the inference from serum PRL to tumor pathway activity is therefore indirect. Routine clinical sampling left residual sources of variation, including nonstandardized sampling time, possible stress effects, medication or physiological influences not fully captured in the record, and undetected macroprolactin. Estrogen-related endocrine differences may also influence circulating PRL (48) and were only partly addressed through adjustment for age and menopausal status. We did not assess psychological factors, such as depression or chronic stress, which have been linked to dysregulation of the hypothalamic-pituitary-adrenal axis and may influence circulating hormone levels, including that of PRL (49). These factors could represent an unmeasured source of variation in our PRL measurements. Moreover, because the circulating level of PRL declines with age and after menopause and because the four groups differed substantially in age, part of the cross-group PRL pattern—including the higher median PRL in the younger HC group than in the older IDC group—may reflect age and reproductive status rather than inflammation. The principal contrasts were therefore either age-matched (NPM vs. HCs) or covariate-adjusted, but residual confounding cannot be excluded. Furthermore, the BMI for the IDC group was computed from recorded height and weight, and menopausal status was harmonized from differently coded source datasets via a prespecified binary mapping. Additionally, the molecular-subtype analysis was limited by small per-subtype sample sizes, especially luminal A (n=8) and HER2-enriched disease (n=4), and the reported predictive values depend on the subtype mix and IBC prevalence of this single-center sample. Along with this, the dichotomy contrasting luminal B HER2+ plus TNBC against the other subtypes was not prespecified but was defined after the subtype distribution had been examined; the corresponding OR and P value should therefore be regarded as hypothesis-generating only, are likely to overstate the true effect, and should be confirmed in an independent, prospectively defined cohort. Finally, the clinical nature of the AJCC T4d criteria means that the IBC group is clinically rather than biologically defined, and observed PRL patterns may reflect the shared inflammatory syndrome with NPM rather than IBC-specific tumor biology. The borderline-IBC sensitivity analysis was conducted with disease-course duration rather than with skin-involvement extent because the source dataset recorded skin involvement only as a binary field.

Conclusions

Serum PRL level was elevated in inflammatory breast conditions but did not distinguish IBC from NPM, indicating that circulating PRL level should be considered as a shared inflammation-associated endocrine signal rather than as a disease-specific diagnostic marker. The separation observed between IBC and NPM was driven chiefly by CRP and age. Within the IBC group, hyperprolactinemia appeared more frequent in luminal B HER2+ and TNBC tumors in an exploratory post hoc analysis; however, given the small sample size (42 IBC patients, 13 hyperprolactinemic cases) and severely limited statistical power, this finding is strictly hypothesis-generating and requires prospective validation in an independent, adequately powered cohort. Because tumor PRLR expression and signaling activation were not assessed, any inference regarding PRL/PRLR pathway activation in IBC tumor cells remains indirect and requires tissue-based confirmation. These findings represent the first description of serum PRL in the IBC population and warrant confirmation in larger, prospective, multicenter cohorts.

Supplementary

The article’s supplementary files as

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. The study was approved by the Ethics Committee of Yueyang Hospital of Integrated Traditional Chinese and Western Medicine, Shanghai University of Traditional Chinese Medicine (approval No. 2026-104). The requirement for written informed consent was waived due to the retrospective nature of the analysis and the use of anonymized routine clinical data.

Data Sharing Statement

Available at https://gs.amegroups.com/article/view/10.21037/gs-2026-0582/dss

gs-15-10-292-dss.pdf ^{(72.3KB, pdf)}

DOI: 10.21037/gs-2026-0582